In 2050, Artificial Intelligence will surpass human intelligence!
This pivotal date has been identified by some specialists as the advent of stage 5 artificial intelligence: self-awareness.
In around 100 years (since Turing's invention), the machine will have gone from being an "extension of the human brain" (stages 1 to 3) to a form of autonomy. Could the next stage be independence?
A utopian vision would say that Man's work must be done for the common good, or at least for the greatest number. The Asimovian vision would be that Man's creation "cannot harm a human being nor, remaining passive, leave that human being exposed to danger". Historical contradictions are tenacious, and it's surely for this reason that the European Union is currently legislating on the "Artificial Intelligence Act".
Even before speculating on the consequences of autonomous, or even independent, artificial intelligence, it is vital to consider the duties and responsibilities of the organizations that develop artificial intelligence and those that use it, particularly for commercial purposes.
In the end, it all comes down to the same issue: whatever the means used to exploit it, my personal data must not be used against me, whatever the purpose, and I must remain the sole owner and decision-maker of this data.
What does the AI Act say about this, at the time of writing:
4 thresholds for the dangerousness of artificial intelligence: minimal or zero, limited, high and unacceptable. The penultimate one concerns the scope of application in healthcare, justice, education, etc.
This concerns all companies developing artificial intelligence, from the moment it interacts within the European Union.
Geometry varies according to whether the AI is open source or closed source: in fact, the constraints would appear to be less stringent in the case of open source. Note that "Llama", published by Meta (Facebook), is considered open source.
So, to sum up, the Artificial Intelligence Act (AI Act) aims to regulate artificial intelligence according to its risks. It's a world first. A first step towards responsible AI.
The Act will come into force in 2026. At the same time, a European AI Office will be set up to ensure compliance with the law, and to impose financial penalties on companies that fail to comply. No doubt it will be one of the vectors of the law's evolution, over time.
Implications for companies:
The use of artificial intelligence will be constrained by the sector of activity and the use cases: an organization in the healthcare field will have many more constraints than a simple retailer.
The issue of protecting and securing personal data will become exponentially more complex (remember: AI is currently in stage 4), especially if the organization has an international dimension. Storm warning for DPOs...
Publishers of solutions based on artificial intelligence will have to keep their commercial promises in line with new regulations and corporate values. Facial recognition at store entrances...in Europe, will have to wait.
To make provisions for risks: the law provides for a system of fines. 35 million euros or 7% of total annual sales, to 7.5 million euros or 1.5% of sales for lesser offences.
Educating employees in the use of AI: what happens with the mass of information, some of which may be sensitive or even critical (without realizing it) that passes through the AI prompts (powerpoint presentation generation, help with the sales pitch for a brand-new product, product image generation, etc.)? All this is data which must also be protected.
Complexity of technical and functional evaluation by procurement teams (or by the IT department by delegation): the incremental and accelerating complexity of AI-powered solutions will make upgrading and maintaining the skills of the teams involved costly and constant. Towards AI cerberus to ensure the responsible use of AI technology?
This post, although based on facts, is the expression of an analysis and an opinion. Let's discuss it, at will, to understand this revolution in the best possible way.
Comments